Together AI's $800M raise and the case for open models

Together AI just raised $800 million at an $8.3 billion valuation to run a cloud built around open-weight models — DeepSeek, MiniMax, Kimi — instead of the closed APIs from OpenAI and Anthropic. The interesting part isn't the check. It's what the check is betting on: that enough businesses want off the closed-model treadmill to build an $8B company serving them.

What actually happened

On July 1, TechCrunch reported that Together AI closed an $800M Series C led by Aramco Ventures, with Nvidia, Vista Equity Partners, General Catalyst, and others participating. The round more than doubles the valuation from its $305M Series B at $3.3B about sixteen months ago.

The pitch to enterprises is blunt: run open-source models for training and inference and pay less than you would for closed systems. Together says annual bookings have topped $1.15 billion, and it plans to scale compute roughly 50x over the next five years. Per its own announcement, the money goes into inference capacity and new products. The through-line the investors are buying: enterprises are moving workloads to open models to control cost and avoid lock-in.

Why it matters for your business

You don't need Together AI specifically. You need the option they represent. When your automation runs on an open-weight model, the model is a file with a documented interface — you can run it on Together, on your own GPUs, on Baseten, or on whatever's cheapest next quarter. When it runs on a closed API, you're renting behavior that can be repriced, rate-limited, deprecated, or gated by a government the week you depend on it. We've watched all four happen this year.

That doesn't mean rip out every closed model — the frontier ones are still better at hard reasoning, and for a lot of jobs that's worth paying for. It means know which of your workloads actually need the frontier and which are running a $30-per-million-token model to do classification a $1 open model handles fine. Put every model call behind one interface in your own code. Keep your prompts, eval sets, and data in your systems, not the vendor's console. Then swapping the engine is a config change, not a rebuild.

We build AI systems the same way regardless of which model wins: the logic and data are yours, the model is a swappable input, and the bill is a line item you can move. Open-model clouds like Together just make the swap cheaper.

Sources: TechCrunch, Together AI / BusinessWire.